{"id":"W2807926734","doi":"10.1145/3209635.3209655","title":"Exam Wrappers","year":2018,"lang":"en","type":"article","venue":"","topic":"Evaluation of Teaching Practices","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; University of Calgary","funders":"","keywords":"Medical education; Computer science; Drop out; Test (biology); Mathematics education; Psychology; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002605131,0.0005676787,0.0004334963,0.001576569,0.0007524248,0.00113088,0.001018467,0.0005480162,0.03357327],"category_scores_gemma":[0.01842849,0.0002635316,0.0007148887,0.0008438706,0.0003306621,0.0007512312,0.001370894,0.0008060014,0.008355285],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004922457,"about_ca_system_score_gemma":0.001578308,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001941376,"about_ca_topic_score_gemma":0.004344364,"domain_scores_codex":[0.9981427,0.0005479809,0.00020862,0.0002159039,0.0006368862,0.0002479199],"domain_scores_gemma":[0.9877993,0.004854779,0.001748575,0.001789346,0.001907539,0.001900514],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0006305883,0.002803054,0.01366382,0.001150385,0.00006425507,0.0004134444,0.001142436,0.0004979678,0.009655908,0.001422927,0.04175597,0.9267993],"study_design_scores_gemma":[0.0005659123,0.008254099,0.2532185,0.002127404,0.0003919851,0.001979311,0.003047709,0.002035076,0.03500249,0.003159512,0.6900012,0.0002167584],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"other","genre_scores_codex":[0.656258,0.004275174,0.08665728,0.003298349,0.001722534,0.007730532,0.006877337,0.01572593,0.2174549],"genre_scores_gemma":[0.6953881,0.00440514,0.1318462,0.001987395,0.0004496741,0.002830772,0.005383539,0.0008798457,0.1568294],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.03357327,"threshold_uncertainty_score":0.1123137,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2508113948073107,"score_gpt":0.5182574327298312,"score_spread":0.2674460379225205,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}